CN108010021A - A kind of magic magiscan and method - Google Patents
A kind of magic magiscan and method Download PDFInfo
- Publication number
- CN108010021A CN108010021A CN201711242641.8A CN201711242641A CN108010021A CN 108010021 A CN108010021 A CN 108010021A CN 201711242641 A CN201711242641 A CN 201711242641A CN 108010021 A CN108010021 A CN 108010021A
- Authority
- CN
- China
- Prior art keywords
- medical image
- target pixel
- pixel points
- pixel
- smart network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The embodiment of the invention discloses a kind of magic magiscan and method.The system includes：One or more processors；Memory, for storing one or more programs, when one or more of programs are performed by one or more of processors so that one or more of processors are realized：Obtain medical image；The medical image is inputted into the first smart network, obtain the first class probability figure of the medical image, every bit in the first class probability figure is corresponding with the pixel of the medical image, and first smart network is advance according to medical image sample and the training acquisition of corresponding target pixel points；Target pixel points are determined in the medical image according to the first class probability figure, the collection of the target pixel points is combined into destination object.The embodiment of the present invention improves the detection rates of target pixel points, ensure that the accuracy rate of destination object detection.
Description
Technical field
The present embodiments relate to Medical Image Processing, more particularly to a kind of magic magiscan and method.
Background technology
When carrying out medical image detection, testing result is usually there are two kinds of situations, and one kind is real target area, i.e.,
The body position of actual lesion；One kind is false positive, i.e., reality do not occur lesion but body position that testing result is lesion.
Thus, when carrying out medical image detection, it is most important for the position for being correctly detecting lesion to filter out false positive.
The existing method to lesion detection, carries out initial survey, then using convolutional neural networks using sliding window first
False positive is carried out to operate.Due to lesion and false positive shared pixel ratio very little in the picture, carried out using sliding window
, there are the problem of missing inspection or flase drop, when the size of window is too small, there is detection when the size of window is too big in the mode of initial survey
The problem of speed is too slow.The accuracy rate for ultimately resulting in lesion detection is low and speed is slow.
The content of the invention
The embodiment of the present invention provides a kind of magic magiscan and method, solves the standard of existing lesion detection method
True rate is low and the problem of speed is slow.
In a first aspect, an embodiment of the present invention provides a kind of magic magiscan, which includes：
One or more processors；
Memory, for storing one or more programs,
When one or more of programs are performed by one or more of processors so that one or more of processing
Device is realized：
Obtain medical image；
The medical image is inputted into the first smart network, obtains the first class probability figure of the medical image,
Every bit in the first class probability figure is corresponding with the pixel of the medical image, first smart network
It is that advance trained according to medical image sample and corresponding target pixel points obtains；
Treat to determine target pixel points, the target pixel points in medical image described according to the first class probability figure
Collection be combined into destination object.
Further, target pixel points bag is determined in the medical image to be detected according to the first class probability figure
Include：
Binary conversion treatment is carried out to the first class probability figure according to predetermined threshold value, obtains binary map；
Isolated connected domain is determined in the binary map, and in the isolated corresponding medical image to be detected of connected domain
Pixel be target pixel points.
Further, the medical image is inputted the first smart network includes：
Medical image is pre-processed, the pretreatment includes at least one in segmentation, filtering or sliding window processing
Kind；
The pretreated medical image is inputted into the first smart network.
Further, the processor is additionally operable to perform：
The medical image that the target pixel points determine is input in the second smart network and obtains the second classification generally
Rate, wherein, second smart network is that advance foundation target sample and definite false positive position are trained；
According to the second class probability of second smart network output, in the set for determining the target pixel points
False positive pixel；
False positive pixel in the set of the target pixel points is removed, obtains medical image testing result.
Further, the processor is additionally operable to perform：
Feature extraction is carried out to the set of the target pixel points, obtains feature extraction result；
The feature extraction result is inputted into grader to determine false positive pixel, and the false positive pixel is gone
Remove, obtain testing result.
Second aspect, the embodiment of the present invention additionally provide a kind of medical image processing method, and this method includes：
Medical image is obtained, the medical image includes multiple pixels；
The medical image is inputted into the first smart network, the pixel of the medical image is categorized as target
Pixel and non-targeted pixel, first smart network are advance according to medical image sample and corresponding target picture
What vegetarian refreshments was trained；
First smart network includes the first output channel and the second output channel, and the target pixel points include
First kind target pixel points and the second class target pixel points, the pixel of first output channel output medical image belong to the
The classification results of a kind of target pixel points, the pixel of the second output channel output medical image belong to the second class target picture
The classification results of vegetarian refreshments.
Further, the collection of the target pixel points is combined into destination object, and the medical image includes lung areas and rib
The CT images or DR images in bone region, the corresponding destination object of set of the first kind target pixel points are the section of lung areas
Point, the corresponding destination object of set of the second class target pixel points are the discontinuous point of rib region.
Further, the medical image is inputted into the first smart network, by the pixel of the medical image
Being categorized as target pixel points and non-targeted pixel includes：
The medical image is inputted into the first smart network, obtains multiple objective contours, first artificial intelligence
Network includes the mapping relations of medical image and objective contour；
Pixel in the objective contour is target pixel points, and the pixel outside the objective contour is non-targeted pixel
Point.
Further, the medical image is inputted into the first smart network, by the pixel of the medical image
Being categorized as target pixel points and non-targeted pixel includes：
The medical image is inputted into the first smart network, obtains the class probability of the pixel of the medical image
Value, first smart network include the mapping relations of the class probability value of medical image and pixel；
The pixel of the medical image is categorized as by target pixel points and non-targeted pixel according to the class probability value
Point.
Further, the method further includes：
Determine the false positive pixel in the medical image, and the vacation is removed from the set of the target pixel points
Positive pixel.
The embodiment of the present invention is trained artificial by using advance foundation medical image sample and corresponding target pixel points
Intelligent network carries out the detection of target pixel points, and since smart network used is end to end network, it has by inputting
The characteristic of output is directly obtained, improves the acquisition speed of pixel probability distribution, and then improves destination object detection rates,
And the position of target pixel points is determined according to probability graph, it is ensured that the accuracy rate of destination object detection.
Brief description of the drawings
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing does one and simply introduces, it should be apparent that, drawings in the following description are some embodiments of the present invention, for this
For the those of ordinary skill of field, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is a kind of structure diagram for magic magiscan that the embodiment of the present invention one provides；
Fig. 2 is a kind of flow chart for medical image processing method that the embodiment of the present invention one provides；
Fig. 3 is a kind of medical image to be detected that the embodiment of the present invention one provides；
Fig. 4 is the binary map for the first class probability figure that the embodiment of the present invention one obtains；
Fig. 5 is a kind of flow chart of medical image processing method provided by Embodiment 2 of the present invention；
Fig. 6 is the schematic network structure of V-net provided by Embodiment 2 of the present invention；
Fig. 7 is a kind of flow chart for medical image processing method that the embodiment of the present invention three provides；
Fig. 8 is the VGG schematic network structures used in a kind of RPN that the embodiment of the present invention three provides；
Fig. 9 a are the CT image schematic diagrames used in the embodiment of the present invention three；
Fig. 9 b are the class probability figure binaryzation result schematic diagram obtained to the CT image procossings of Fig. 9 a；
Figure 10 a are another CT images schematic diagram of the embodiment of the present invention three；
Figure 10 b are the class probability figure binaryzation result schematic diagram obtained to the CT image procossings of Figure 10 a.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, hereinafter with reference to attached in the embodiment of the present invention
Figure, technical scheme is clearly and completely described by embodiment, it is clear that described embodiment is the present invention one
Section Example, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not doing
Go out all other embodiments obtained under the premise of creative work, belong to the scope of protection of the invention.
Embodiment one
The embodiment of the present invention one proposes a kind of magic magiscan, which can be used for reality
Apply the ad hoc approach disclosed in some embodiments of the invention.Particular system in the present embodiment illustrates one using functional block diagram
Hardware platform comprising display module.In certain embodiments, magic magiscan can pass through its hardware device, software
Program, firmware and combinations thereof realize the specific implementation of some embodiments of the invention.In certain embodiments, medicine figure
As processing system can be the computer of a general purpose, or a medical imaging devices with image processing function.Doctor
It can be electrocardiograph (electrocardiography), digital radiography (digital to learn imaging device
Radiography, DR) equipment, magnetic resonance imaging (magnetic resonance imaging, MRI) equipment, computerized tomography
Scan (computed tomography, CT) equipment, PET-Positron emission computed tomography imaging (positron emission
Computed tomography, PET) equipment, ultrasound (ultrasonic, US) equipment, fluorescence spectrum (fluorescence) set
Standby, Single Photon Emission Computed fault imaging (single photon emission computed tomography,
SPECT), the single mode equipment such as mammary machine (mammo) equipment, can also be comprising above two or plurality of devices imaging function
Multimodal device.
As shown in Figure 1, magic magiscan 10 can include internal communication bus 101, processor (processor)
102nd, read-only storage (ROM) 103, random access memory (RAM) 104, communication port 105, input output assembly 106, hard
Disk 107 and user interface 108.Internal communication bus 101 can realize the data communication of different inter-modules.Processor 102 can
To be judged and be sent prompting.In certain embodiments, processor 102 can be made of one or more processors, wherein
Processor may include end to end network composition artificial intelligence (artificial intelligence, AI) network, by medicine
Image, which inputs the smart network, can automatically determine the position of lesion on the medical image.Certainly, it is artificial in processor
Intelligent network was trained by a large amount of priori datas, and network parameters at different levels are constantly adjusted in the training process.Alternatively,
Smart network can include the first smart network and the second smart network, and the preferred end of smart network is arrived
Hold network.Communication port 105 can realize magic magiscan 10 with miscellaneous part for example：External scanner, image are adopted
Collect between equipment, database, external storage and image processing workstations etc. into row data communication.
In certain embodiments, magic magiscan 10 can be sent and received by communication port 105 from network
Information and data.Input output assembly 106 supports the input/output number between magic magiscan 10 and miscellaneous part
According to stream.User interface 108 can realize the interaction and information exchange between magic magiscan 10 and user.Medical image
Processing system 10 can also include various forms of program storage units and data storage element, such as hard disk 107, read-only to deposit
Reservoir (ROM) 103, random access memory (RAM) 104, can store computer disposal and/or the various data that use of communication
File, and the possible programmed instruction performed by processor 102.In some instances, in magic magiscan 10
Program storage unit and data storage element can further comprise relative to the remotely located memory of processor 102, these
Remote memory can pass through network connection to equipment.The example of above-mentioned network include but not limited to internet, intranet,
LAN, mobile radio communication and combinations thereof.
In certain embodiments, input output assembly 106 can be used for the numeral or character information for receiving input, and production
The raw key signals input related with the user setting and function control of equipment/terminal/server.User interface 108 may include
The display devices such as display screen.In certain embodiments, medical image can be lung CT image, by Medical Image Processing side
Method, can detect from medical image and obtain pulmonary nodule region, the region or fracture of rib region that pulmonary emphysema occur；Can be with
Detect that polyp of colon detects CT surprise attacks point automatically, the cerebral hemorrhage of CT images detects automatically, above-mentioned lesions position all can be in user circle
Identified on face 108.In certain embodiments, medical image can be the image of mammary machine shooting, be examined automatically for mammary gland block
Survey or Breast Calcifications detect automatically, fine quantized result can also be shown in user interface 108.
Fig. 2 be the embodiment of the present invention one provide a kind of medical image processing method flow chart, the finger that this method is related to
Order can be performed by processor 102.The technical solution of the present embodiment can be adapted for being detected the lesion of medical image.The party
Method specifically includes following operation：
S210, obtain medical image, which may include multiple pixels.Alternatively, medical image to be detected
It can be the instantaneous acquiring from the scanner that medical image system 10 connects, be sent after reconstruction into processor 102；Can also
It is to be stored in advance in hard disk 107, read-only storage (ROM) 103 or random access memory (RAM) 104 and be transferred to processing
In device 102.The type of medical image can be one-dimensional (1D) data, two-dimentional (2D) image or three-dimensional (3D) image.For example, 1D
Data can be electrocardiogram；2D images can be radioscopic image；3D rendering can be MR images, CT images, PET image, ultrasound
One or more combinations in image, fluoroscopic image, SPECT images.Exemplarily, medical image can correspond to person under inspection's
Lung areas, mammary region, colon regions, head zone etc..It is alternatively possible to image intensity, image are carried out to medical image
The normalizeds such as contrast, picture format, image slice spacing operate.Alternatively, the form of medical image can be DICOM
Form, binary format or NIFTI forms.
Medical image, is inputted the first smart network by S220, and the first classification of acquisition each pixel of medical image is general
Rate figure.Wherein, first smart network is advance according to medical image sample and corresponding target pixel points (or target
The probability distribution of pixel) training generation, the every bit in the first class probability figure is corresponding with the pixel of medical image,
Processing of each pixel Jing Guo neutral net i.e. in medical image can obtain each pixel and belong to target pixel points
Probable value or probability distribution.
First smart network can be in advance according to a large amount of medical image samples and corresponding definite lesions position
End-to-end (end-to-end) network of generation, end to end network refer to that input is initial data, and output can be two passages
Or multichannel, without extracting feature, autonomous learning feature is convenient and efficient.
The medical image is inputted the first smart network to may also include：
Medical image is pre-processed, the pretreatment includes segmentation, filtering or sliding window processing etc..
Common dividing method has the dividing method based on threshold value, the dividing method based on edge, is based in image segmentation
The dividing method in region etc..The texture of lesion, form etc. are characterized in being different from normal structure.The image is split
In a width medical image, the characteristics of being different from normal structure using lesion, candidate's lesions position, candidate's lesions position are partitioned into
That is area-of-interest.Image segmentation, which can also be, comes out organ segmentation interested, for example, lung, mammary gland, colon or brain
Deng.
The method of image filtering has histogram equalization, gaussian filtering, medium filtering or mean filter etc..Image filtering
Method, which can be realized, filters out noise, and the candidate's lesion ensured is more accurate.
Initial detecting is carried out using sliding window, sliding window can be one-dimensional, two-dimentional or three-dimensional sliding window, for carrying
Take candidate's lesion.Sliding window can cover view picture medical image to be detected, it is ensured that obtained candidate's lesion it is comprehensive.
The pretreated medical image is inputted into the first smart network.Using preprocess method to medical image
Carry out processing and reduce the data volume being input in the first smart network, improve lesion detection rates, pass through pretreatment
The combination of method and end to end network, pretreatment are used as Preliminary detection, and end to end network optimizes, and it is true to improve lesions position
Fixed accuracy rate.
S230, according to the first class probability figure in the medical image determine target pixel points, the collection of target pixel points
Conjunction can be destination object.
Target pixel points can be the pixel of the pixel of focal area or any area-of-interest.First
The first probability graph that smart network produces can reflect that each pixel is the probability or each picture of target pixel points
Element belongs to the probability of non-targeted pixel.For example, for CT lung images, pass through the probability point of all pixels to entire image
Analysis, it may be determined that the node in lung images, the node can be pulmonary nodule regions.In another example can be according to all general
Rate value given threshold is split, and the segmentation figure after segmentation is extracted/determined isolated area, and each isolated area is target
The set of pixel composition.Further, destination object can determine that according to the form of isolated area, destination object can be lung qi
Swollen region, the discontinuous point (breakpoint) in rib cage or the region that pneumothorax occurs.
Determine that target pixel points include in the medical image according to the first class probability figure：
Binary conversion treatment is carried out to the first class probability figure according to predetermined threshold value, obtains binary map；
Predetermined threshold value can be obtained according to the probability Data-Statistics in the first probability graph.According to predetermined threshold value to the first probability graph
It can be that the pixel value that will be greater than predetermined threshold value is set to 1 to carry out binary conversion treatment, will be set to 0 less than the pixel value of predetermined threshold value,
It is achieved in binaryzation.
Isolated connected domain is determined in the binary map, and in the isolated corresponding medical image to be detected of connected domain
Pixel be target pixel points.
Isolated connected domain, that is, pixel value in binary map is all higher than predetermined threshold value or the respectively less than pixel value of predetermined threshold value, and
And surrounding pixel is opposite with the characteristic of the pixel value of pixel in isolated connected region.Lesion detection is improved using binaryzation mode
Speed, it is more accurate with independent communication domain characterization lesions position
Alternatively, isolated area is extracted in the medical image, individual 2D probability graph successively stacks to form 3D probability graphs, pin
To 3D probability graph threshold applications, the mask of binaryzation is obtained.In this binaryzation mask, extraction in the range of 3D is obtained into every pixel
The region that all marks for being connected are in 26 contiguous ranges.
Fig. 3 shows the medical image to be detected in an embodiment, which is CT lung images, it includes focal area
With various puncta vasculosas.According to the above method the two of the first class probability figure as shown in Figure 4 is finally obtained using end to end network
Two isolated connected domains of the higher pixel composition of value figure, wherein brightness correspond to the knuckle areas of CT lungs.
The embodiment of the present invention is generated by using advance foundation medical image sample and corresponding pixel probability distribution
End to end network carries out the detection of target pixel points, due to the characteristic that output is directly obtained by input of end to end network, improves
The acquisition speed of pixel probability distribution, and then improve lesion detection rates, and lesions position is determined according to probability graph
It can ensure the accuracy rate of lesion detection；The processing speed of end to end network is fast, and without splicing to result images, precision is more
Height, can reduce post-processing operation, processing speed is faster.
Embodiment two
Fig. 5 is a kind of flow chart of medical image processing method provided by Embodiment 2 of the present invention, the finger that this method is related to
Order can equally be performed by processor 102.The technical solution of the present embodiment can be adapted for being detected the lesion of medical image,
And further lesion is screened to remove false positive.This method specifically includes following operation：
S510, obtain medical image, which may include multiple pixels.In this embodiment, the medical image
For CT lung images, and initial segmentation also is carried out to the medical image, obtain lung areas.
Medical image, is inputted the first smart network by S520, and the first classification of acquisition each pixel of medical image is general
Rate figure.
S530, according to the first class probability figure in the medical image determine target pixel points.
S540, removal target pixel points form the false positive pixel in set, obtain testing result.In an optional reality
Apply in mode, remove the false positive pixel in target pixel points formation set, obtaining testing result includes：
A1, by the medical image that the target pixel points determine be input in the second smart network obtain second classification
Probability, wherein, second smart network is that advance trained according to target sample and definite false positive position obtains
's.
A2, the second class probability exported according to second smart network, determine the collection of the target pixel points
False positive pixel in conjunction.
A3, by the target pixel points formed set in false positive pixel remove, obtain medical image testing result.
In this particular embodiment, the false positive in candidate's lesion and true lesion can be distinguished using 2D/3D-CNN
Come, the first smart network can use 2D-CNN, i.e., carry out convolution operation using 2D convolution kernels；Second smart network
3D-CNN can be used, i.e., convolution operation is carried out using 3D convolution kernels.
Alternatively, the first smart network or the second smart network may include：Full convolutional neural networks, U-net
Or V-net.Wherein, full convolutional neural networks (FCN, Fully Convolutional Network) are to each in image
Pixel is classified, and reaches the effect classified to image specific part with this.U-net, is the upgrade of network of FCN, will roll up
Intermediate result during product is added in the computing of transposition convolution.
Fig. 6 is the schematic network structure of V-net, and the left-hand component of above-mentioned network is used to extract characteristics of image, the right portion
Divide the feature group for obtaining extraction to merge the probability graph for being extended to original image size, specifically refer to Milletari F,
Navab N,Ahmadi S A.V-net:Fully convolutional neural networks for volumetric
medical image segmentation[C]//3D Vision(3DV),2016Fourth International
Conference on.IEEE,2016:565-571.The intermediate result that the V-net is remained in the convolution process of U-net is added to
This feature in transposition convolution algorithm, and Dice functions have been used as object function, and used residual error network
The skip floor of (residual net).Dice functions can calculate the similarity of two objects：
Wherein, piBelong to the voxel of prediction object；giFor the voxel of actual object；1≤i≤N, N are voxel number, and N is
Integer more than 1.
Destination object sample is priori focal area, and the second smart network is advance according to focal area and definite
False positive position trains to have obtained parameter matrix, then the medical image that target pixel points are determined is input to trained second people
Work intelligent network is i.e. available with inputting corresponding i.e. the second probability graph of false positive probability figure, can be true by the second probability graph
Determine false positive position.Corresponding position in focal area can be deleted according to false positive position, obtain the position of final lesion.
Using the further selective mechanisms of the second smart network as a result, improving the accuracy rate of lesion detection.
In another optional embodiment, the false positive pixel in the set of target pixel points is removed, is detected
As a result include：
B1, the set formed using feature extracting method to the target pixel points carry out feature extraction, obtain feature and carry
Take result.
Wherein, the feature extracting method includes at least：Utilize statistics feature, morphological feature, textural characteristics or small
The method that wave characteristic carries out feature extraction.
B2, by the feature extraction result input grader to determine false positive, and the false positive is removed, obtained most
Whole lesion testing result.
Wherein, grader can be support vector machines (SVM, Support Vector Machine), linear classifier
(LDA, Linear Discriminant Analysis) or neutral net, can use convolutional neural networks in neutral net
(CNN, Convolutional Neural Network), exemplarily, two dimension or three-dimensional volume can be used for 3-D view
Product neutral net is classified.
Detect citing automatically with Lung neoplasm.Examined using RPN networks as the network of initial survey, detection for each CT scan
Measure to one group of boundary rectangle, each boundary rectangle and represent a lesion, i.e. candidate's Lung neoplasm.In all candidate's Lung neoplasms,
Include several real Lung neoplasms, and a large amount of false positives.Therefore, second step operation is carried out, removes false positive.To use up can
Removal these false positives more than energy, leave real Lung neoplasm.Split network by lesion used here as V-net, Unet, FCN etc.
Network is inputted one by one, exports the segmentation result of these tubercles.If false positive, then export one all close to zero probability
Figure.If real tubercle, then tubercle corresponding pixel points have high probability in the probability graph of output.Added on probability graph
One threshold value, then obtain the profile of tubercle.If the point of any high probability is not produced, then it is assumed that this candidate nodule is false sun
Property.Export the candidate nodule containing high probability point, and its profile.Medical image is carried out by using the first smart network
Initial survey, determines the position of lesion, and the operation of false positive is further carried out using feature extraction and the combination of grader, is realized
The high accuracy detection of lesions position.
Embodiment three
Fig. 7 is a kind of flow chart for medical image processing method that the embodiment of the present invention three provides.The technology of the present embodiment
Scheme is further optimized on the basis of above-mentioned any embodiment.The method of the present embodiment includes：
S710, obtain medical image, and the medical image includes multiple pixels.
S720, by the medical image input the first smart network, by each pixel of the medical image point
Class is target pixel points and non-targeted pixel, and first smart network is advance according to medical image sample and correspondence
Target pixel points training obtain.
In this embodiment, first smart network's selection region suggests network (region proposal
Network, RPN) medical image inputted into the first smart network, each pixel of the medical image is classified
Include for target pixel points and non-targeted pixel：The medical image is inputted into RPN networks, obtains multiple objective contours, institute
Stating RPN networks includes the mapping relations of medical image and objective contour；Pixel in the objective contour is target pixel points,
Pixel outside the objective contour is non-targeted pixel.Fig. 8 is the VGG network knots used in the embodiment of the present invention in RPN
Structure schematic diagram.The network includes successively：Two convolutional layers, pond layer, two convolutional layers, pond layer, three convolutional layers, Chi Hua
Layer, three convolutional layers, cost layers.By the exportable boundary rectangle of above-mentioned network (objective contour) and the probability of target pixel points.
The concrete structure of above-mentioned network also refers to：Ren S,He K,Girshick R,et al.Faster R-CNN:Towards
real-time object detection with region proposal networks[C]//Advances in
neural information processing systems.2015:91-99。
Above-mentioned zone suggests one kind that network is end to end network, and the end to end network includes the first output channel and the
Two output channels, the target pixel points include first kind target pixel points and the second class target pixel points, first output
The pixel of passage output medical image belongs to the classification results of first kind target pixel points, the second output channel output doctor
The pixel for learning image belongs to the classification results of the second class target pixel points.
Export and exported for two passages or multichannel, two passages, refer to one passage of prospect, one passage of background.Prospect refers to
Lesion.If first kind target pixel points are prospect, i.e. lesion, the second class target pixel points are background.Multichannel result refers to
The result of multiple classification.Such as：Detect a variety of lesions at the same time.Specifically, such as：In chest CT detect Lung neoplasm, pulmonary emphysema,
Fracture of rib etc., and such as：Detected in x-ray chest radiograph, while detect tumour, pneumothorax, fracture etc..At this moment three detections pair
As plus a background, then result figure can export 4 passages.
Fig. 9 a are the CT image schematic diagrames used in the embodiment of the present invention, wherein, the place presence of (arrow) is identified in figure
Knuckle areas, i.e. destination object, other parts also have similar knuckle areas, but only can not accurately recognize reality by naked eyes
The specific location of border lesion.Fig. 9 b are the class probability figure binaryzation result (i.e. binary map) obtained to the CT image procossings of Fig. 9 a
Schematic diagram, the overwhelming majority is normal structure (the relatively low region of brightness) in figure, and the obvious high brightness shown in the picture
Region is the knuckle areas of corresponding CT.
Figure 10 a are another CT images schematic diagram of the embodiment of the present invention, and according to priori, which is normal structure
Image, right side identification division be vascular tissue in figure, and the region is easy to be misjudged after using the first smart network detection
For focal area.Figure 10 b are class probability figure binaryzation result (binary map) schematic diagram obtained to the CT images of Figure 10 a, are schemed
The obvious high-brightness region shown as in corresponds to the angiosomes of CT, rather than focal area.Detection knot is improved in order to obtain
The accuracy of fruit, in another preferred embodiment of the present embodiment, the medical image processing method further includes：
Determine the false positive pixel in the medical image, and the false positive picture is removed from the target pixel points
Vegetarian refreshments.The removal of false positive pixel refers to the method in embodiment two.
Note that it above are only presently preferred embodiments of the present invention and institute's application technology principle.It will be appreciated by those skilled in the art that
The invention is not restricted to specific embodiment described here, can carry out for a person skilled in the art various obvious changes,
Readjust and substitute without departing from protection scope of the present invention.Therefore, although being carried out by above example to the present invention
It is described in further detail, but the present invention is not limited only to above example, without departing from the inventive concept, also
It can include other more equivalent embodiments, and the scope of the present invention is determined by scope of the appended claims.
Claims (10)
1. a kind of magic magiscan, the system comprises：
One or more processors；
Memory, for storing one or more programs,
When one or more of programs are performed by one or more of processors so that one or more of processors are real
It is existing：
Obtain medical image；
The medical image is inputted into the first smart network, obtains the first class probability figure of the medical image, it is described
Every bit in first class probability figure is corresponding with the pixel of the medical image, and first smart network is pre-
First obtained according to medical image sample and the training of corresponding target pixel points；
Target pixel points, the set of the target pixel points are determined in the medical image according to the first class probability figure
For destination object.
2. system according to claim 1, it is characterised in that according to the first class probability figure in the medical image
In determine target pixel points include：
Binary conversion treatment is carried out to the first class probability figure according to predetermined threshold value, obtains binary map；
Isolated connected domain is determined in the binary map, with the picture in the isolated corresponding medical image to be detected of connected domain
Vegetarian refreshments is target pixel points.
3. system according to claim 1, it is characterised in that the medical image is inputted into the first smart network bag
Include：
Medical image is pre-processed, the pretreatment includes at least one of segmentation, filtering or sliding window processing；
The pretreated medical image is inputted into the first smart network.
4. system according to claim 1, it is characterised in that the processor is additionally operable to perform：
The medical image that the target pixel points determine is input in the second smart network and obtains the second class probability, its
In, second smart network is that advance foundation target sample and definite false positive position are trained；
The second class probability exported according to second smart network, determines the vacation in the set of the target pixel points
Positive pixel；
False positive pixel in the set of the target pixel points is removed, obtains medical image testing result.
5. system according to claim 1, it is characterised in that the processor is additionally operable to perform：
Feature extraction is carried out to the set of the target pixel points, obtains feature extraction result；
The feature extraction result is inputted into grader to determine false positive pixel, and the false positive pixel is removed,
Obtain testing result.
6. a kind of medical image processing method, the described method includes：
Medical image is obtained, the medical image includes multiple pixels；
The medical image is inputted into the first smart network, the pixel of the medical image is categorized as object pixel
Point and non-targeted pixel, first smart network are advance according to medical image sample and corresponding target pixel points
What training obtained；
First smart network includes the first output channel and the second output channel, and the target pixel points include first
Class target pixel points and the second class target pixel points, the pixel of the first output channel output medical image belong to the first kind
The classification results of target pixel points, the pixel of the second output channel output medical image belong to the second class target pixel points
Classification results.
It is 7. described according to the method described in claim 6, it is characterized in that, the collection of the target pixel points is combined into destination object
Medical image includes the CT images or DR images of lung areas and rib region, and the set of the first kind target pixel points corresponds to
Destination object be lung areas node, the corresponding destination object of set of the second class target pixel points is rib region
Discontinuous point.
8. according to the method described in claim 6, it is characterized in that, by the medical image input the first smart network,
Included so that the pixel of the medical image is categorized as target pixel points and non-targeted pixel：
The medical image is inputted into the first smart network, obtains multiple objective contours, first smart network
Mapping relations including medical image and objective contour；
Pixel in the objective contour is target pixel points, and the pixel outside the objective contour is non-targeted pixel.
9. according to the method described in claim 6, it is characterized in that, by the medical image input the first smart network,
Included so that the pixel of the medical image is categorized as target pixel points and non-targeted pixel：
The medical image is inputted into the first smart network, obtains the class probability value of the pixel of the medical image,
First smart network includes the mapping relations of the class probability value of medical image and pixel；
The pixel of the medical image is categorized as by target pixel points and non-targeted pixel according to the class probability value.
10. according to the method described in claim 6, it is characterized in that, the method further includes：
Determine the false positive pixel in the medical image, and the false positive is removed from the set of the target pixel points
Pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711242641.8A CN108010021B (en) | 2017-11-30 | 2017-11-30 | Medical image processing system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711242641.8A CN108010021B (en) | 2017-11-30 | 2017-11-30 | Medical image processing system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108010021A true CN108010021A (en) | 2018-05-08 |
| CN108010021B CN108010021B (en) | 2021-12-10 |
Family
ID=62055669
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711242641.8A Active CN108010021B (en) | 2017-11-30 | 2017-11-30 | Medical image processing system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108010021B (en) |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108682015A (en) * | 2018-05-28 | 2018-10-19 | 科大讯飞股份有限公司 | Lesion segmentation method, apparatus, equipment and storage medium in a kind of biometric image |
| CN109117703A (en) * | 2018-06-13 | 2019-01-01 | 中山大学中山眼科中心 | It is a kind of that cell category identification method is mixed based on fine granularity identification |
| CN109272510A (en) * | 2018-07-24 | 2019-01-25 | 清华大学 | The dividing method of tubular structure in a kind of 3 d medical images |
| CN109447973A (en) * | 2018-10-31 | 2019-03-08 | 腾讯科技（深圳）有限公司 | A kind for the treatment of method and apparatus and system of polyp of colon image |
| CN109671049A (en) * | 2018-11-07 | 2019-04-23 | 哈尔滨工业大学（深圳） | A kind of medical image processing method, system, equipment, storage medium |
| CN109685807A (en) * | 2018-11-16 | 2019-04-26 | 广州市番禺区中心医院(广州市番禺区人民医院、广州市番禺区心血管疾病研究所) | Lower-limb deep veins thrombus automatic division method and system based on deep learning |
| CN110009598A (en) * | 2018-11-26 | 2019-07-12 | 腾讯科技（深圳）有限公司 | Method and image segmentation apparatus for image segmentation |
| CN110221346A (en) * | 2019-07-08 | 2019-09-10 | 西南石油大学 | A kind of data noise drawing method based on the full convolutional neural networks of residual block |
| CN110349161A (en) * | 2019-07-10 | 2019-10-18 | 北京字节跳动网络技术有限公司 | Image partition method, device, electronic equipment and storage medium |
| CN110555833A (en) * | 2019-08-30 | 2019-12-10 | 联想(北京)有限公司 | image processing method, image processing apparatus, electronic device, and medium |
| CN110599467A (en) * | 2019-08-29 | 2019-12-20 | 上海联影智能医疗科技有限公司 | Method and device for detecting non-beam limiter area, computer equipment and storage medium |
| WO2020000643A1 (en) * | 2018-06-28 | 2020-01-02 | 深圳视见医疗科技有限公司 | Method and device for detecting pulmonary nodule in ct image, and readable storage medium |
| WO2020001086A1 (en) * | 2018-06-28 | 2020-01-02 | 深圳先进技术研究院 | Deep learning-based cardiac ct image segmentation method and apparatus, device, and medium |
| CN110689947A (en) * | 2018-07-04 | 2020-01-14 | 天津天堰科技股份有限公司 | Display device and display method |
| CN110706217A (en) * | 2019-09-26 | 2020-01-17 | 中国石油大学(华东) | Deep learning-based lung tumor automatic delineation method |
| CN110766659A (en) * | 2019-09-24 | 2020-02-07 | 西人马帝言(北京)科技有限公司 | Medical image recognition method, apparatus, device and medium |
| CN110838114A (en) * | 2019-11-11 | 2020-02-25 | 苏州锐一仪器科技有限公司 | Pulmonary nodule detection method, device and computer storage medium |
| CN111275707A (en) * | 2020-03-13 | 2020-06-12 | 北京深睿博联科技有限责任公司 | Pneumonia focus segmentation method and device |
| WO2020238817A1 (en) * | 2019-05-24 | 2020-12-03 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for processing x-ray images |
| CN109544560B (en) * | 2018-10-31 | 2021-04-27 | 上海商汤智能科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN112819811A (en) * | 2021-02-24 | 2021-05-18 | 上海商汤智能科技有限公司 | Image analysis method and related device, electronic equipment and storage medium |
| CN112950600A (en) * | 2021-03-10 | 2021-06-11 | 武汉联影智融医疗科技有限公司 | Brain identification extraction method and device, computer equipment and storage medium |
| CN113096109A (en) * | 2021-04-22 | 2021-07-09 | 四川大学 | Lung medical image analysis method, device and system |
| CN113256634A (en) * | 2021-07-13 | 2021-08-13 | 杭州医策科技有限公司 | Cervical carcinoma TCT slice vagina arranging method and system based on deep learning |
| CN113940701A (en) * | 2021-10-11 | 2022-01-18 | 什维新智医疗科技(上海)有限公司 | Trachea and blood vessel detection device based on thyroid gland |
| CN114612971A (en) * | 2022-03-04 | 2022-06-10 | 北京百度网讯科技有限公司 | Face detection method, model training method, electronic device, and program product |
| CN115272206A (en) * | 2022-07-18 | 2022-11-01 | 深圳市医未医疗科技有限公司 | Medical image processing method, medical image processing device, computer equipment and storage medium |
| CN116452791A (en) * | 2023-03-27 | 2023-07-18 | 广州市斯睿特智能科技有限公司 | Multi-camera point defect area positioning method, system, device and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020969A (en) * | 2012-12-25 | 2013-04-03 | 中国科学院深圳先进技术研究院 | Processing method and processing system for dividing liver graphs of CT (computed tomography) image |
| CN105389821A (en) * | 2015-11-20 | 2016-03-09 | 重庆邮电大学 | Medical image segmentation method based on combination of cloud module and image segmentation |
| CN105574859A (en) * | 2015-12-14 | 2016-05-11 | 中国科学院深圳先进技术研究院 | Liver tumor segmentation method and device based on CT (Computed Tomography) image |
| CN105894517A (en) * | 2016-04-22 | 2016-08-24 | 北京理工大学 | CT image liver segmentation method and system based on characteristic learning |
| CN105957066A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on automatic context model |
| CN106056596A (en) * | 2015-11-30 | 2016-10-26 | 浙江德尚韵兴图像科技有限公司 | Fully-automatic three-dimensional liver segmentation method based on local apriori information and convex optimization |
| CN106940816A (en) * | 2017-03-22 | 2017-07-11 | 杭州健培科技有限公司 | Connect the CT image Lung neoplasm detecting systems of convolutional neural networks entirely based on 3D |
| CN107256550A (en) * | 2017-06-06 | 2017-10-17 | 电子科技大学 | A kind of retinal image segmentation method based on efficient CNN CRF networks |
| CN107292887A (en) * | 2017-06-20 | 2017-10-24 | 电子科技大学 | A kind of Segmentation Method of Retinal Blood Vessels based on deep learning adaptive weighting |
-
2017
- 2017-11-30 CN CN201711242641.8A patent/CN108010021B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020969A (en) * | 2012-12-25 | 2013-04-03 | 中国科学院深圳先进技术研究院 | Processing method and processing system for dividing liver graphs of CT (computed tomography) image |
| CN105389821A (en) * | 2015-11-20 | 2016-03-09 | 重庆邮电大学 | Medical image segmentation method based on combination of cloud module and image segmentation |
| CN106056596A (en) * | 2015-11-30 | 2016-10-26 | 浙江德尚韵兴图像科技有限公司 | Fully-automatic three-dimensional liver segmentation method based on local apriori information and convex optimization |
| CN105574859A (en) * | 2015-12-14 | 2016-05-11 | 中国科学院深圳先进技术研究院 | Liver tumor segmentation method and device based on CT (Computed Tomography) image |
| CN105894517A (en) * | 2016-04-22 | 2016-08-24 | 北京理工大学 | CT image liver segmentation method and system based on characteristic learning |
| CN105957066A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on automatic context model |
| CN106940816A (en) * | 2017-03-22 | 2017-07-11 | 杭州健培科技有限公司 | Connect the CT image Lung neoplasm detecting systems of convolutional neural networks entirely based on 3D |
| CN107256550A (en) * | 2017-06-06 | 2017-10-17 | 电子科技大学 | A kind of retinal image segmentation method based on efficient CNN CRF networks |
| CN107292887A (en) * | 2017-06-20 | 2017-10-24 | 电子科技大学 | A kind of Segmentation Method of Retinal Blood Vessels based on deep learning adaptive weighting |
Cited By (41)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108682015A (en) * | 2018-05-28 | 2018-10-19 | 科大讯飞股份有限公司 | Lesion segmentation method, apparatus, equipment and storage medium in a kind of biometric image |
| CN108682015B (en) * | 2018-05-28 | 2021-10-19 | 安徽科大讯飞医疗信息技术有限公司 | Focus segmentation method, device, equipment and storage medium in biological image |
| CN109117703A (en) * | 2018-06-13 | 2019-01-01 | 中山大学中山眼科中心 | It is a kind of that cell category identification method is mixed based on fine granularity identification |
| CN109117703B (en) * | 2018-06-13 | 2022-03-22 | 中山大学中山眼科中心 | Hybrid cell type identification method based on fine-grained identification |
| WO2020001086A1 (en) * | 2018-06-28 | 2020-01-02 | 深圳先进技术研究院 | Deep learning-based cardiac ct image segmentation method and apparatus, device, and medium |
| US10909682B2 (en) | 2018-06-28 | 2021-02-02 | Shenzhen Imsight Medical Technology Co. Ltd. | Method and device for detecting pulmonary nodule in computed tomography image, and computer-readable storage medium |
| WO2020000643A1 (en) * | 2018-06-28 | 2020-01-02 | 深圳视见医疗科技有限公司 | Method and device for detecting pulmonary nodule in ct image, and readable storage medium |
| CN110689947A (en) * | 2018-07-04 | 2020-01-14 | 天津天堰科技股份有限公司 | Display device and display method |
| CN109272510A (en) * | 2018-07-24 | 2019-01-25 | 清华大学 | The dividing method of tubular structure in a kind of 3 d medical images |
| CN109272510B (en) * | 2018-07-24 | 2021-06-04 | 清华大学 | Method for segmenting tubular structure in three-dimensional medical image |
| CN109447973A (en) * | 2018-10-31 | 2019-03-08 | 腾讯科技（深圳）有限公司 | A kind for the treatment of method and apparatus and system of polyp of colon image |
| CN109544560B (en) * | 2018-10-31 | 2021-04-27 | 上海商汤智能科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN109447973B (en) * | 2018-10-31 | 2021-11-26 | 腾讯医疗健康(深圳)有限公司 | Method, device and system for processing colon polyp image |
| CN109671049B (en) * | 2018-11-07 | 2024-03-01 | 哈尔滨工业大学（深圳） | Medical image processing method, system, equipment and storage medium |
| CN109671049A (en) * | 2018-11-07 | 2019-04-23 | 哈尔滨工业大学（深圳） | A kind of medical image processing method, system, equipment, storage medium |
| CN109685807A (en) * | 2018-11-16 | 2019-04-26 | 广州市番禺区中心医院(广州市番禺区人民医院、广州市番禺区心血管疾病研究所) | Lower-limb deep veins thrombus automatic division method and system based on deep learning |
| CN110009598B (en) * | 2018-11-26 | 2023-09-05 | 腾讯科技（深圳）有限公司 | Method for image segmentation and image segmentation device |
| CN110009598A (en) * | 2018-11-26 | 2019-07-12 | 腾讯科技（深圳）有限公司 | Method and image segmentation apparatus for image segmentation |
| EP3958742A4 (en) * | 2019-05-24 | 2022-06-29 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for processing x-ray images |
| WO2020238817A1 (en) * | 2019-05-24 | 2020-12-03 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for processing x-ray images |
| CN110221346A (en) * | 2019-07-08 | 2019-09-10 | 西南石油大学 | A kind of data noise drawing method based on the full convolutional neural networks of residual block |
| CN110349161A (en) * | 2019-07-10 | 2019-10-18 | 北京字节跳动网络技术有限公司 | Image partition method, device, electronic equipment and storage medium |
| CN110349161B (en) * | 2019-07-10 | 2021-11-23 | 北京字节跳动网络技术有限公司 | Image segmentation method, image segmentation device, electronic equipment and storage medium |
| CN110599467A (en) * | 2019-08-29 | 2019-12-20 | 上海联影智能医疗科技有限公司 | Method and device for detecting non-beam limiter area, computer equipment and storage medium |
| CN110555833A (en) * | 2019-08-30 | 2019-12-10 | 联想(北京)有限公司 | image processing method, image processing apparatus, electronic device, and medium |
| CN110766659A (en) * | 2019-09-24 | 2020-02-07 | 西人马帝言(北京)科技有限公司 | Medical image recognition method, apparatus, device and medium |
| CN110706217A (en) * | 2019-09-26 | 2020-01-17 | 中国石油大学(华东) | Deep learning-based lung tumor automatic delineation method |
| CN110838114A (en) * | 2019-11-11 | 2020-02-25 | 苏州锐一仪器科技有限公司 | Pulmonary nodule detection method, device and computer storage medium |
| CN110838114B (en) * | 2019-11-11 | 2022-07-22 | 苏州锐一仪器科技有限公司 | Pulmonary nodule detection method, device and computer storage medium |
| CN111275707B (en) * | 2020-03-13 | 2023-08-25 | 北京深睿博联科技有限责任公司 | Pneumonia focus segmentation method and device |
| CN111275707A (en) * | 2020-03-13 | 2020-06-12 | 北京深睿博联科技有限责任公司 | Pneumonia focus segmentation method and device |
| CN112819811A (en) * | 2021-02-24 | 2021-05-18 | 上海商汤智能科技有限公司 | Image analysis method and related device, electronic equipment and storage medium |
| CN112950600A (en) * | 2021-03-10 | 2021-06-11 | 武汉联影智融医疗科技有限公司 | Brain identification extraction method and device, computer equipment and storage medium |
| CN113096109A (en) * | 2021-04-22 | 2021-07-09 | 四川大学 | Lung medical image analysis method, device and system |
| CN113256634B (en) * | 2021-07-13 | 2021-11-23 | 杭州医策科技有限公司 | Cervical carcinoma TCT slice vagina arranging method and system based on deep learning |
| CN113256634A (en) * | 2021-07-13 | 2021-08-13 | 杭州医策科技有限公司 | Cervical carcinoma TCT slice vagina arranging method and system based on deep learning |
| CN113940701A (en) * | 2021-10-11 | 2022-01-18 | 什维新智医疗科技(上海)有限公司 | Trachea and blood vessel detection device based on thyroid gland |
| CN114612971A (en) * | 2022-03-04 | 2022-06-10 | 北京百度网讯科技有限公司 | Face detection method, model training method, electronic device, and program product |
| CN115272206A (en) * | 2022-07-18 | 2022-11-01 | 深圳市医未医疗科技有限公司 | Medical image processing method, medical image processing device, computer equipment and storage medium |
| CN116452791A (en) * | 2023-03-27 | 2023-07-18 | 广州市斯睿特智能科技有限公司 | Multi-camera point defect area positioning method, system, device and storage medium |
| CN116452791B (en) * | 2023-03-27 | 2024-03-22 | 广州市斯睿特智能科技有限公司 | Multi-camera point defect area positioning method, system, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108010021B (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108010021A (en) | A kind of magic magiscan and method | |
| Jiang et al. | Multiple resolution residually connected feature streams for automatic lung tumor segmentation from CT images | |
| CN105559813B (en) | Medical diagnostic imaging apparatus and medical image-processing apparatus | |
| CN110310281A (en) | Lung neoplasm detection and dividing method in a kind of Virtual Medical based on Mask-RCNN deep learning | |
| CN107622492A (en) | Lung splits dividing method and system | |
| Zhong et al. | Boosting‐based cascaded convolutional neural networks for the segmentation of CT organs‐at‐risk in nasopharyngeal carcinoma | |
| JP2023507109A (en) | Automated tumor identification and segmentation from medical images | |
| CN103501699A (en) | Method and apparatus for isolating a potential anomaly in imaging data and its application to medical imagery | |
| CN108062749B (en) | Identification method and device for levator ani fissure hole and electronic equipment | |
| CN109300136B (en) | Automatic segmentation method for organs at risk based on convolutional neural network | |
| Palma et al. | Detection of masses and architectural distortions in digital breast tomosynthesis images using fuzzy and a contrario approaches | |
| Tan et al. | Analysis of segmentation of lung parenchyma based on deep learning methods | |
| Hussain et al. | Femur segmentation in DXA imaging using a machine learning decision tree | |
| Mughal et al. | Adaptive hysteresis thresholding segmentation technique for localizing the breast masses in the curve stitching domain | |
| Rani et al. | Superpixel with nanoscale imaging and boosted deep convolutional neural network concept for lung tumor classification | |
| Heydarheydari et al. | Auto-segmentation of head and neck tumors in positron emission tomography images using non-local means and morphological frameworks | |
| Sornapudi et al. | EpithNet: Deep regression for epithelium segmentation in cervical histology images | |
| Green et al. | 3-D Neural denoising for low-dose Coronary CT Angiography (CCTA) | |
| Ramasamy et al. | Machine learning in cyber physical systems for healthcare: brain tumor classification from MRI using transfer learning framework | |
| Mridha et al. | A comprehensive survey on the progress, process, and challenges of lung cancer detection and classification | |
| Tummala et al. | Liver tumor segmentation from computed tomography images using multiscale residual dilated encoder‐decoder network | |
| Gong et al. | Towards accurate pulmonary nodule detection by representing nodules as points with high-resolution network | |
| Xu et al. | Improved cascade R-CNN for medical images of pulmonary nodules detection combining dilated HRNet | |
| Dhalia Sweetlin et al. | Patient-Specific Model Based Segmentation of Lung Computed Tomographic Images. | |
| Jalab et al. | Fractional Renyi entropy image enhancement for deep segmentation of kidney MRI |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant after: Shanghai Lianying Medical Technology Co., Ltd Address before: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant before: SHANGHAI UNITED IMAGING HEALTHCARE Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |